Detecting salient locations or reliable feature points in a visual scene has been a research subject for decades, both in computational neuroscience and classical computer vision. Because visual saliency is closely related to human perception, many models incorporate models from biological visual perception. In particular, since the response properties of neurons in the retina and primary visual cortex have been intensively studied and best understood, computational operations associated with these brain areas have been widely incorporated into saliency models. However, processing associated with higher cortical areas is only beginning to be incorporated, partly due to a lack of models and experimental results....[ Read more ]

Detecting salient locations or reliable feature points in a visual scene has been a research subject for decades, both in computational neuroscience and classical computer vision. Because visual saliency is closely related to human perception, many models incorporate models from biological visual perception. In particular, since the response properties of neurons in the retina and primary visual cortex have been intensively studied and best understood, computational operations associated with these brain areas have been widely incorporated into saliency models. However, processing associated with higher cortical areas is only beginning to be incorporated, partly due to a lack of models and experimental results.

We describe a salient point detection model that incorporates recent findings about the response properties of neurons in cortical area V2. It has been reported that certain V2 neurons encode combinations of orientations, with an apparent inclination to orthogonal pairs. The model introduced here integrates this finding as the last stage of a hierarchical architecture modeled after the visual processing in mammalian brains. Most parts of this model are based on current knowledge of neuroscience, and the seemingly simple operations like linear filtering can lead to fair resemblance to response properties of different neurons. We demonstrate that this model captures the intuitive notion of saliency, detects repeatable feature points under various image transformations and noisy conditions. We also introduce a binarization method and discuss the advantages of employing this method in our model.